When we started Replicate in 2019, OpenAI had just open sourced GPT-2, and few people outside the machine learning community paid much attention to AI. But those of us who were on the ground felt as if something big was about to happen. Remarkable models were being built in academic laboratories, but you needed a metaphorical lab coat to be able to run them.
We’ve made it our mission to get research models out of the lab and into the hands of developers. We wanted programmers to creatively bend and twist these models into products that researchers might never have thought of.
We looked at it as a tooling issue. In the same way that tools like Heroku made it possible to run a website without managing a web server, we wanted to create tools to run models without having to understand backpropagation or deal with CUDA errors.
the first device we made Teeth: A standard packaging format for machine learning models. then we made repeat As a platform for running Cog models as API endpoints in the cloud. We’ve taken away both the low-level machine learning and complex GPU cluster management you need to make inferences at scale.
It turns out the timing was just right. When? stable spread Released in 2022, we had mature infrastructure that could handle the massive developer interest in running these models. Lots of great apps and products were built on Replicat, apps that often ran a single model packaged in a sleek UI to solve a particular use case.
Since then, AI engineering Has matured into a serious craft. AI apps are no longer limited to just running models. The modern AI stack has model inference, but also microservices, content delivery, object storage, caching, databases, telemetry, etc. We see many of our customers building complex heterogeneous stacks, where replicated models are a part of higher-order systems across multiple platforms.
This is why we’re joining CloudflareReplicate contains the tools and primitives to run models, Cloudflare has the best network, workers, R2, durable objects, and all the other primitives you need to build a complete AI stack,
The AI stack lives entirely on the network. The models run on data center GPUs and are glued together by small cloud functions that call vector databases, fetch objects from blob storage, call MCP servers, etc.network computer is“This has never been truer.
At Cloudflare, we will now be able to build the AI infrastructure layer we have dreamed of from the beginning. We’ll be able to do things like run agile models at the edge, run model pipelines on instant-booting workers, stream model input and output with WebRTC, and more.
We’re proud of what we’ve built at Replika. We were the first generic AI services platform, and we defined the abstractions and design patterns that most of our peers have adopted. We have developed an amazing community of builders and researchers around our product.
<a href